Mobility-Aware Resource Allocation in D2D Communications Using Genetic Algorithms

Abstract

Smart resource allocation techniques are needed to meet the increasing need for low-latency and high-throughput communication in 5G and beyond networks. Direct user-to-user communication provided by Device-to-Device (D2D) communication increases spectrum efficiency but also makes power control, interference management, and mobility adaptation more difficult. Under dynamic user mobility, static allocation strategies frequently fail, resulting in poor performance and unequal resource distribution. This paper offers a machine learning (ML) based resource allocation framework guaranteeing fairness and minimizing interference that uses mobility prediction and a multi-objective Non-Dominated Sorting Genetic Algorithm 2 (NSGA-II) to optimize power control, channel assignment, and spectrum reuse in D2D-enabled cellular networks. The suggested method demonstrates through simulation a 25% improvement in throughput, a reduction in interference, and significant enhancement in Jain's fairness index compared to baseline approaches. Higher Quality of Service (QoS) in dynamic settings depends on this performance improvement. The framework also offers insightful analysis of how mobility affects Signal to Interference plus Noise Ratio (SINR), power efficiency, and QoS, stressing the need for dynamic adaptability in real-time systems. While multiple works on resource optimization already exist, including those using deep learning (DL) and federated learning (FL) for mobility-aware ML optimization, our contribution lies in integrating zone-based mobility prediction with multi-objective optimization. Our approach provides a practical basis for future research, including deep reinforcement learning, federated learning, and large-scale real-world deployments.